[Bugfix] Restore `weight_dtype` in `QuarkW8A8Fp8MoEMethod` to fix GPT-OSS FP8 MoE weight loading (#54824)
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
Micah Williamson committed
Sep 1, 2026 at 16:08 UTC
ce6a283c097cd573f6f371bbb376c4e5c4d07ef7
1 file changed
+1
vllm/model_executor/layers/quantization/quark/quark_moe.py
+1
@@ -243,6 +243,7 @@ class QuarkW8A8Fp8MoEMethod(QuarkMoEMethod):
243
activation_quant_key: QuantKey | None,
244
):
245
super().__init__(moe, weight_quant_key, activation_quant_key)
246
+ self.weight_dtype = "fp8"
247
self.weight_qscheme = (
248
"per_channel" if weight_quant_key == kFp8StaticChannelSym else "per_tensor"
249
)